jinql
/
bashtop
mirror of https://github.com/aristocratos/bashtop
1
0
Fork 0
Code Issues Releases Wiki Activity

Added GPUtil error checking

pull/237/head
energizerbee 2023-08-11 19:40:38 -07:00
parent 0463d08cbc
commit 55ebc229c1
2 changed files with 87 additions and 38 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
README.md
View File

@ -62,7 +62,7 @@ GPUtil uses nvidia-smi as a dependency, the package may not work with some graph
For GPU collection, the python packages psutil, and GPUtil are required as well as nvidia-smi. For GPU collection, the python packages psutil, and GPUtil are required as well as nvidia-smi.
>Install nvidia-smi (Debian/Ubuntu based see installation section for other distros) >Install nvidia-smi (Debian/Ubuntu based see installation section for installation on other distro or OS)
``` bash ``` bash
sudo apt-get install nvidia-smi sudo apt-get install nvidia-smi
@ -412,6 +412,9 @@ hires_graphs="false"
#* Enable the use of psutil python3 module for data collection, default on OSX #* Enable the use of psutil python3 module for data collection, default on OSX
use_psutil="true" use_psutil="true"
#* Enable the use of GPUtil python3 module for data collection, may not work on some cards (namely AMD cards)
use_gputil="true"
``` ```
#### Command line options: (not yet implemented) #### Command line options: (not yet implemented)

120
bashtop
View File

@ -173,7 +173,7 @@ hires_graphs="false"
use_psutil="true" use_psutil="true"
#* Enable the use of GPUtil python3 module for data collection, may not work on some cards (namely AMD cards) #* Enable the use of GPUtil python3 module for data collection, may not work on some cards (namely AMD cards)
use_GPUtil="true" use_gputil="true"
aaz_config() { : ; } #! Do not remove this line! aaz_config() { : ; } #! Do not remove this line!
#? End default variables--------------------------------------------------------------------------------> #? End default variables-------------------------------------------------------------------------------->
@ -210,7 +210,7 @@ declare -A cpu mem gpu swap proc net box theme disks
declare -a cpu_usage gpu_usage cpu_graph_a cpu_graph_b gpu_graph_a gpu_graph_b color_meter color_temp_graph color_cpu color_cpu_graph cpu_history color_gpu_graph color_mem_graph color_swap_graph declare -a cpu_usage gpu_usage cpu_graph_a cpu_graph_b gpu_graph_a gpu_graph_b color_meter color_temp_graph color_cpu color_cpu_graph cpu_history color_gpu_graph color_mem_graph color_swap_graph
declare -a mem_history swap_history gpu_history net_history_download net_history_upload mem_graph gpu_graph swap_graph proc_array download_graph upload_graph trace_array declare -a mem_history swap_history gpu_history net_history_download net_history_upload mem_graph gpu_graph swap_graph proc_array download_graph upload_graph trace_array
declare resized=1 size_error clock tty_width tty_height hex="16#" cpu_p_box swap_on=1 draw_out esc_character boxes_out last_screen clock_out update_string declare resized=1 size_error clock tty_width tty_height hex="16#" cpu_p_box swap_on=1 draw_out esc_character boxes_out last_screen clock_out update_string
declare -a options_array=("color_theme" "update_ms" "use_psutil" "use_GPUtil" "proc_sorting" "proc_tree" "check_temp" "draw_clock" "background_update" "custom_cpu_name" declare -a options_array=("color_theme" "update_ms" "use_psutil" "use_gputil" "proc_sorting" "proc_tree" "check_temp" "draw_clock" "background_update" "custom_cpu_name"
"proc_per_core" "proc_reversed" "proc_gradient" "disks_filter" "hires_graphs" "net_totals_reset" "update_check" "error_logging") "proc_per_core" "proc_reversed" "proc_gradient" "disks_filter" "hires_graphs" "net_totals_reset" "update_check" "error_logging")
declare -a save_array=(${options_array[*]/net_totals_reset/}) declare -a save_array=(${options_array[*]/net_totals_reset/})
declare -a sorting=( "pid" "program" "arguments" "threads" "user" "memory" "cpu lazy" "cpu responsive") declare -a sorting=( "pid" "program" "arguments" "threads" "user" "memory" "cpu lazy" "cpu responsive")
@ -219,7 +219,7 @@ declare -A pid_history
declare time_left timestamp_start timestamp_end timestamp_input_start timestamp_input_end time_string mem_out proc_misc prev_screen pause_screen filter input_to_filter declare time_left timestamp_start timestamp_end timestamp_input_start timestamp_input_end time_string mem_out proc_misc prev_screen pause_screen filter input_to_filter
declare no_epoch proc_det proc_misc2 sleeping=0 detail_mem_graph proc_det2 proc_out curled git_version has_iostat sensor_comm failed_pipes=0 py_error declare no_epoch proc_det proc_misc2 sleeping=0 detail_mem_graph proc_det2 proc_out curled git_version has_iostat sensor_comm failed_pipes=0 py_error
declare esc_character tab backspace sleepy late_update skip_process_draw winches quitting theme_int notifier saved_stty nic_int net_misc skip_net_draw declare esc_character tab backspace sleepy late_update skip_process_draw winches quitting theme_int notifier saved_stty nic_int net_misc skip_net_draw
declare psutil_disk_fail declare psutil_disk_fail gputil_missing
declare -a disks_free disks_total disks_name disks_free_percent saved_key themes nic_list old_procs declare -a disks_free disks_total disks_name disks_free_percent saved_key themes nic_list old_procs
printf -v esc_character "\u1b" printf -v esc_character "\u1b"
printf -v tab "\u09" printf -v tab "\u09"
@ -360,6 +360,10 @@ init_() { #? Collect needed information and set options before startig main loop
print -bg "#00" -fg "#30ff50" -r 1 -t "√" print -bg "#00" -fg "#30ff50" -r 1 -t "√"
print -m $(( (tty_height/2-3)+stx++ )) 0 -bg "#00" -fg "#cc" -c -b "Checking gpu..." print -m $(( (tty_height/2-3)+stx++ )) 0 -bg "#00" -fg "#cc" -c -b "Checking gpu..."
get_gpu_info get_gpu_info
check_gputil_status
if [[ $gputil_missing == "true" ]]; then
print -m $(( (tty_height/2-3)+stx++ )) 0 -bg "#00" -fg "#fa1e1e" -c -b "WARNING: GPUTIL NOT FOUND"
fi
#* Set graph resolution #* Set graph resolution
graph[hires]="${hires_graphs}" graph[hires]="${hires_graphs}"
@ -916,8 +920,14 @@ get_cpu_info() {
fi fi
} }
check_gputil_status(){
local status
py_command -v status "get_gputil_status()"
if [[ $status == "0" ]]; then use_gputil="false"; gputil_missing="true"; fi
}
get_gpu_info(){ get_gpu_info(){
if [[ $use_GPUtil == true ]]; then if [[ $use_gputil == true ]]; then
py_command -v gpu[model] "get_gpu_name()" py_command -v gpu[model] "get_gpu_name()"
py_command -v gpu[driver] "get_gpu_driver()" py_command -v gpu[driver] "get_gpu_driver()"
fi fi
@ -2323,7 +2333,7 @@ collect_mem() { #? Collect memory information from "/proc/meminfo"
} }
collect_gpu() { #? Collect memory and usage information from GPUtil collect_gpu() { #? Collect memory and usage information from GPUtil
if [[ $use_GPUtil == true ]]; then if [[ $use_gputil == true ]]; then
local -a available=("gpu") local -a available=("gpu")
local pygpuout local pygpuout
@ -2949,7 +2959,7 @@ calc_sizes() { #? Calculate width and height of all boxes
done done
#* Copy numbers around to get target layout #* Copy numbers around to get target layout
if [[ $use_GPUtil == true ]]; then if [[ $use_gputil == true ]]; then
#* Make space for the gpu box if it is being used #* Make space for the gpu box if it is being used
box[gpu_line]=$((box[mem_line])) box[gpu_line]=$((box[mem_line]))
box[gpu_col]=$((box[processes_col])) box[gpu_col]=$((box[processes_col]))
@ -3010,7 +3020,7 @@ calc_sizes() { #? Calculate width and height of all boxes
box[n_col]="$((net_width-box[n_width]+2))" box[n_col]="$((net_width-box[n_width]+2))"
box[n_line]="$((net_line+(net_height/2)-(box[n_height]/2)+1))" box[n_line]="$((net_line+(net_height/2)-(box[n_height]/2)+1))"
if [[ $use_GPUtil == true ]]; then if [[ $use_gputil == true ]]; then
#* Calculate placement of gpu value box #* Calculate placement of gpu value box
local gpu_line=$((box[gpu_line]+1)) gpu_width=$((box[gpu_width]-2)) gpu_height=$((box[gpu_height]-2)) local gpu_line=$((box[gpu_line]+1)) gpu_width=$((box[gpu_width]-2)) gpu_height=$((box[gpu_height]-2))
box[g_width]=32 box[g_width]=32
@ -3025,7 +3035,7 @@ draw_bg() { #? Draw all box outlines
unset boxes_out unset boxes_out
for this_box in ${box[boxes]}; do for this_box in ${box[boxes]}; do
if [[ $this_box == "gpu" && $use_GPUtil == false ]]; then continue; fi if [[ $this_box == "gpu" && $use_gputil == false ]]; then continue; fi
create_box -v boxes_out -col ${box[${this_box}_col]} -line ${box[${this_box}_line]} -width ${box[${this_box}_width]} -height ${box[${this_box}_height]} -fill -lc "${box[${this_box}_color]}" -title ${this_box} create_box -v boxes_out -col ${box[${this_box}_col]} -line ${box[${this_box}_line]} -width ${box[${this_box}_width]} -height ${box[${this_box}_height]} -fill -lc "${box[${this_box}_color]}" -title ${this_box}
done done
@ -3049,7 +3059,7 @@ draw_bg() { #? Draw all box outlines
print -v boxes_out -m $((box[n_line]+box[n_height]-2)) $((box[n_col]+1)) -rs -fg ${theme[div_line]} -t "┤" -fg ${theme[title]} -b -t "Upload" -rs -fg ${theme[div_line]} -t "├" print -v boxes_out -m $((box[n_line]+box[n_height]-2)) $((box[n_col]+1)) -rs -fg ${theme[div_line]} -t "┤" -fg ${theme[title]} -b -t "Upload" -rs -fg ${theme[div_line]} -t "├"
#* Misc gpu box #* Misc gpu box
if [[ $use_GPUtil == true ]]; then if [[ $use_gputil == true ]]; then
gpu_model_len=${#gpu[model]} gpu_model_len=${#gpu[model]}
gpu_driver_len=${#gpu[driver]} gpu_driver_len=${#gpu[driver]}
create_box -v boxes_out -col $((box[g_col]-1)) -line $((box[g_line]-1)) -width ${box[g_width]} -height ${box[g_height]} -lc ${theme[div_line]} -t "${gpu[model]:0:${gpu_model_len}}" create_box -v boxes_out -col $((box[g_col]-1)) -line $((box[g_line]-1)) -width ${box[g_width]} -height ${box[g_height]} -lc ${theme[div_line]} -t "${gpu[model]:0:${gpu_model_len}}"
@ -3297,8 +3307,8 @@ draw_mem() { #? Draw mem, swap and disk statistics
} }
draw_gpu() { #? Draw GPU usage, memory, and temperature graphs draw_gpu() { #? Draw GPU usage, memory, and temperature graphs
if [[ $use_GPUtil == true ]]; then if [[ $use_gputil == true ]]; then
local gpu_out local gpu_out meter_width usage_meter_width values="usage total used free"
#* Get variables from previous calculations #* Get variables from previous calculations
local col=$((box[gpu_col]+1)) line=$((box[gpu_line]+1)) width=$((box[gpu_width]-2)) height=$((box[gpu_height]-2)) local col=$((box[gpu_col]+1)) line=$((box[gpu_line]+1)) width=$((box[gpu_width]-2)) height=$((box[gpu_height]-2))
@ -3318,7 +3328,6 @@ draw_gpu() { #? Draw GPU usage, memory, and temperature graphs
((resized++)) ((resized++))
fi fi
#* I SWEAR I WILL MAKE THIS CLEANER SOON
#* Create graphs and meters for the gpu box #* Create graphs and meters for the gpu box
create_graph -o gpu_graph_a -d ${line} ${col} ${graph_a_size} $((width-p_width-box[g_width]-1)) -c color_gpu_graph -n gpu_history create_graph -o gpu_graph_a -d ${line} ${col} ${graph_a_size} $((width-p_width-box[g_width]-1)) -c color_gpu_graph -n gpu_history
create_graph -o gpu_graph_b -d $((line+graph_a_size)) ${col} ${graph_b_size} $((width-p_width-box[g_width]-1)) -c color_gpu_graph -i -n gpu_history create_graph -o gpu_graph_b -d $((line+graph_a_size)) ${col} ${graph_b_size} $((width-p_width-box[g_width]-1)) -c color_gpu_graph -i -n gpu_history
@ -3329,32 +3338,56 @@ draw_gpu() { #? Draw GPU usage, memory, and temperature graphs
core_name="cpu_core_1_graph" core_name="cpu_core_1_graph"
meter="${!core_name}" meter="${!core_name}"
create_meter -v gpu_usage_meter -w $((box[g_width]-2)) -f -c color_gpu_graph $((gpu[usage])) for value in ${values}; do
create_meter -v gpu_used_meter -w $((box[g_width]-6)) -f -c color_used_graph ${gpu[used_percent]} if [[ $value == "total" ]]; then continue; fi
create_meter -v gpu_free_meter -w $((box[g_width]-6)) -f -c color_free_graph ${gpu[free_percent]} if (($g_height>9)) then
meter_width=$((box[g_width]-6))
usage_meter_width=$((box[g_width]-2))
elif (($g_height>7)) then
meter_width=12
usage_meter_width=$((box[g_width]-2))
else
meter_width=12
usage_meter_width=15
fi
if [[ $value == "usage" ]]; then create_meter -v gpu_usage_meter -w $usage_meter_width -f -c color_gpu_graph $((gpu[usage])); continue; fi
create_meter -v gpu_${value}_meter -w $meter_width -f -c color_${value}_graph ${gpu[${value}_percent]}
done
#* Humanize values that include memory #* Humanize memory values
local ypos=$g_line local ypos=$g_line
floating_humanizer -v gpu[total] -s 1 -B "${gpu[total]}" for value in ${values}; do
floating_humanizer -v gpu[used] -s 1 -B "${gpu[used]}" if [[ $value == "usage" ]]; then continue; fi
floating_humanizer -v gpu[free] -s 1 -B "${gpu[free]}" floating_humanizer -v gpu[${value}] -s 1 -B "${gpu[${value}]}"
done
#* Draw the labels, meters, and values in the gpu box #* Draw the labels, meters, and values in the gpu box depending on size
print -v gpu_out -rs -fg ${theme[title]} -b -m $((ypos++)) $g_col -jl 10 -t "GPU Usage:"\ if (($g_height>7)); then
-rs -fg ${main_fg} -jr 20 -t "$((gpu[usage]))%" -m $((ypos++)) $((box[g_col])) -t $gpu_usage_meter -rs -fg $normal_color print -v gpu_out -rs -fg ${theme[title]} -b -m $((ypos++)) $g_col -jl 10 -t "GPU Usage:"\
$((ypos++)) -rs -fg ${main_fg} -jr 20 -t "$((gpu[usage]))%" -m $((ypos++)) $((box[g_col])) -t $gpu_usage_meter -rs -fg $normal_color
if (($g_height>11)); then ((ypos++)); fi
else
print -v gpu_out -rs -fg ${theme[title]} -b -m $((ypos++)) $g_col -jl 11 -t "GPU Usage:" -t $gpu_usage_meter -rs -fg ${main_fg} -jr 4 -t "$((gpu[usage]))%" -rs -fg $normal_color
fi
if (($g_height>9)); then
print -v gpu_out -rs -fg ${theme[title]} -b -m $((ypos++)) $g_col -jl 10 -t "GPU Memory:" -jr 19 -t "${gpu[total]}"
print -v gpu_out -rs -fg ${main_fg} -m $((ypos++)) $g_col -jl 10 -t "Used:" -jr 20 -t "${gpu[used]}"\
-m $((ypos++)) $((box[g_col])) -t $gpu_used_meter -rs -fg $normal_color -rs -fg ${main_fg} -jr 4 -t "${gpu[used_percent]}%"
elif (($g_height>6)); then
print -v gpu_out -rs -fg ${theme[title]} -b -m $((ypos++)) $g_col -jl 10 -t "GPU Memory:" -jr 19 -t "${gpu[total]}"
print -v gpu_out -rs -fg ${main_fg} -m $((ypos++)) $g_col -jl 6 -t "Used:" -jl 4 -t "${gpu[used_percent]}%" -t $gpu_used_meter -rs -fg $normal_color -rs -fg ${main_fg} -jr 8 -t "${gpu[used]}"
fi
print -v gpu_out -rs -fg ${theme[title]} -b -m $((ypos++)) $g_col -jl 10 -t "GPU Memory:" -jr 19 -t "${gpu[total]}" if (($g_height>9)); then
print -v gpu_out -rs -fg ${main_fg} -m $((ypos++)) $g_col -jl 10 -t "Free:" -jr 20 -t "${gpu[free]}"\
-m $((ypos++)) $((box[g_col])) -t $gpu_free_meter -rs -fg $normal_color -rs -fg ${main_fg} -jr 4 -t "${gpu[free_percent]}%"
if (($g_height>11)) then ((ypos++)); fi
elif (($g_height>6)); then
print -v gpu_out -rs -fg ${main_fg} -m $((ypos++)) $g_col -jl 6 -t "Free:" -jl 4 -t "${gpu[free_percent]}%" -t $gpu_free_meter -rs -fg $normal_color -rs -fg ${main_fg} -jr 8 -t "${gpu[free]}"
fi
print -v gpu_out -rs -fg ${main_fg} -m $((ypos++)) $g_col -jl 10 -t "Used:" -jr 20 -t "${gpu[used]}"\ print -v gpu_out -m $((ypos++)) $g_col -rs -fg ${theme[title]} -jl 14 -b -t "Temperature:" -rs -fg ${theme[inactive_fg]} "⡀⡀⡀⡀⡀⡀⡀⡀⡀⡀" -l 10 -fg $gpu_temp_color -t "$meter"\
-m $((ypos++)) $((box[g_col])) -t $gpu_used_meter -rs -fg $normal_color -rs -fg ${main_fg} -jr 4 -t "${gpu[used_percent]}%"
print -v gpu_out -rs -fg ${main_fg} -m $((ypos++)) $g_col -jl 10 -t "Free:" -jr 20 -t "${gpu[free]}"\
-m $((ypos++)) $((box[g_col])) -t $gpu_free_meter -rs -fg $normal_color -rs -fg ${main_fg} -jr 4 -t "${gpu[free_percent]}%"
$((ypos++))
print -v gpu_out -m $((ypos++)) $g_col -rs -fg ${theme[main_fg]} -jl 14 -b -t "Temperature:" -rs -fg ${theme[inactive_fg]} "⡀⡀⡀⡀⡀⡀⡀⡀⡀⡀" -l 10 -fg $gpu_temp_color -t "$meter"\
-jr 4 -fg $gpu_temp_color -t "${gpu[temp]}" -fg ${theme[main_fg]} -t "°C" -jr 4 -fg $gpu_temp_color -t "${gpu[temp]}" -fg ${theme[main_fg]} -t "°C"
draw_out+="${gpu_graph_a[*]}${gpu_graph_b[*]}${gpu_graph_temp[*]}${gpu_out}" draw_out+="${gpu_graph_a[*]}${gpu_graph_b[*]}${gpu_graph_temp[*]}${gpu_out}"
@ -3998,7 +4031,7 @@ options_() { #? Shows the options overlay
"True or false." "True or false."
" " " "
"Can only be switched off when on Linux.") "Can only be switched off when on Linux.")
desc_use_GPUtil=( "Enable the use of GPUtil python3 module for" desc_use_gputil=( "Enable the use of GPUtil python3 module for"
"data collection. MAY NOT WORK ON SOME CARDS" "data collection. MAY NOT WORK ON SOME CARDS"
"(namely AMD cards). psutil must be enabled." "(namely AMD cards). psutil must be enabled."
"" ""
@ -4283,7 +4316,7 @@ options_() { #? Shows the options overlay
if ((net[reset]==1)); then net_totals_reset="Off"; net[reset]=0 if ((net[reset]==1)); then net_totals_reset="Off"; net[reset]=0
else net_totals_reset="On"; net[reset]=1; fi else net_totals_reset="On"; net[reset]=1; fi
;; ;;
"check_temp"*|"error_logging"*|"background_update"*|"proc_reversed"*|"proc_gradient"*|"proc_per_core"*|"update_check"*|"hires_graphs"*|"use_psutil"*|"use_GPUtil"*|"proc_tree"*) "check_temp"*|"error_logging"*|"background_update"*|"proc_reversed"*|"proc_gradient"*|"proc_per_core"*|"update_check"*|"hires_graphs"*|"use_psutil"*|"use_gputil"*|"proc_tree"*)
local -n selected_var=${selected} local -n selected_var=${selected}
if [[ ${selected_var} == "true" ]]; then if [[ ${selected_var} == "true" ]]; then
selected_var="false" selected_var="false"
@ -4314,10 +4347,10 @@ options_() { #? Shows the options overlay
fi fi
if [[ $selected == "use_psutil" && $system != "Linux" ]]; then use_psutil="true" if [[ $selected == "use_psutil" && $system != "Linux" ]]; then use_psutil="true"
elif [[ $selected == "use_psutil" ]]; then elif [[ $selected == "use_psutil" ]]; then
use_GPUtil="false" use_gputil="false"
quit_ restart psutil quit_ restart psutil
fi fi
if [[ $selected == "use_GPUtil" ]]; then quit_ restart GPUtil; fi if [[ $selected == "use_gputil" ]]; then quit_ restart GPUtil; fi
if [[ $selected == "error_logging" ]]; then quit_ restart; fi if [[ $selected == "error_logging" ]]; then quit_ restart; fi
;; ;;
@ -5001,11 +5034,18 @@ if [[ $use_psutil == true ]]; then
pywrapper="${pytmpdir}/bashtop.psutil" pywrapper="${pytmpdir}/bashtop.psutil"
cat << 'EOF' > "${pywrapper}" cat << 'EOF' > "${pywrapper}"
import os, sys, subprocess, re, time, psutil, GPUtil import os, sys, subprocess, re, time, psutil
from datetime import timedelta from datetime import timedelta
from collections import defaultdict from collections import defaultdict
from typing import List, Set, Dict, Tuple, Optional, Union from typing import List, Set, Dict, Tuple, Optional, Union
'''Check to see if GPUtil exists'''
gputil_status = 1
try:
import GPUtil
except:
gputil_status = 0
system: str system: str
if "linux" in sys.platform: system = "Linux" if "linux" in sys.platform: system = "Linux"
elif "bsd" in sys.platform: system = "BSD" elif "bsd" in sys.platform: system = "BSD"
@ -5033,6 +5073,7 @@ allowed_commands: Tuple[str] = (
'get_sensors', 'get_sensors',
'get_sensors_check', 'get_sensors_check',
'get_ms', 'get_ms',
'get_gputil_status',
'get_gpu_name', 'get_gpu_name',
'get_gpu_driver', 'get_gpu_driver',
'get_gpu_mem', 'get_gpu_mem',
@ -5163,6 +5204,11 @@ def get_mem():
cmem = mem.active>>10 cmem = mem.active>>10
print(mem.total>>10, mem.free>>10, mem.available>>10, cmem, swap.total>>10, swap.free>>10) print(mem.total>>10, mem.free>>10, mem.available>>10, cmem, swap.total>>10, swap.free>>10)
def get_gputil_status():
'''Check if GPUtil exists'''
status = str(gputil_status)
print(status)
def get_gpu_name(): def get_gpu_name():
'''Fetch GPU model name and chop it to fit in the box''' '''Fetch GPU model name and chop it to fit in the box'''
gpu = GPUtil.getGPUs()[0] gpu = GPUtil.getGPUs()[0]